Pick a tool. Everything runs in your browser and your video stays on your device.
Auto captions run Whisper on your device. Internet is needed once to download the model. For Urdu drama, choose Urdu and Better or Best.
Speech-rich and emotional moments score higher than plain loud music. Very large files are scanned by sampling points across the video, which takes a few minutes. You can stop anytime.
Smart reframe first scans the short for faces and scene cuts, then keeps one steady crop per shot and hard-cuts to whoever is speaking, like an edited drama clip, instead of panning. The preview below stays centered. The scan adds a few seconds before recording.
“Dialogue only” transcribes each short with Whisper and cuts the gaps between spoken lines, so it works even when music plays under the pauses. Pick the right speech language below for best results. First use downloads the speech model.
Checks sample frames for static logos, sponsor banners and still title cards, and replaces those shorts with other picks. It cannot recognise a brand product that appears inside a normal scene.
Step 1: analysis scans the video for its strongest moments. Long episodes take a few minutes; you can stop anytime.
Checks sample frames for static logos, sponsor banners and still title cards, and replaces those shorts with other picks. It cannot recognise a brand product that appears inside a normal scene.
Choose a file from your device. Nothing is uploaded anywhere.
Speech activity, emotional peaks and strong openings are scored second by second.
The top moments become 9:16 clips. Save each one to your device.
Every short gets a score and tags like “Strong hook” or “Dialogue”.
Picks come from the whole video, and “Different picks” re-rolls them.
The 9:16 window follows faces and snaps on scene cuts.
Your video never leaves the device. Only the optional AI models are downloaded.
Clips are recorded in real time from the player, so a 30 second short takes about 30 seconds. Keep the page open and the screen on while exporting.
Older versions ranked by raw loudness, which favors theme music and intros. Now speech activity and peaks relative to the scene around them are used, intro and credits are skipped, and picks are spread across the video. Tap “Different picks” for a new set.
It is signal analysis: speech activity, loudness peaks, motion and opening strength. It finds energetic moments, not meaning, so review the plan before exporting.
Before a short is recorded, a small face detector (downloaded once) scans it for faces and scene cuts. Each shot then gets one steady 9:16 crop that follows the speaker only when they really move, and it jumps exactly on cuts. If the detector cannot load, it falls back to motion.
No. Browsers cannot download YouTube videos, and its terms do not allow it. Download your own video from YouTube Studio and choose the file. Direct .mp4 links work if the site allows it.
A Whisper speech model runs inside your browser, only on the shorts you pick. The first run downloads the model. Urdu accuracy is much better with the Better or Best model. For large files, the short's audio is captured while the clip plays once, so it adds the clip length to prep time. In Export all, the next short's captions are prepared while the current one records.
While a short is recorded, ClipForge jumps over quiet stretches, so the result is tighter than the chosen length. Aggressive removes shorter pauses.
Latest Chrome or Edge (v126 or newer) on Android and desktop, or Safari. Shorts are always saved as MP4. Firefox cannot record MP4, so it is not supported.